Mitigation of human cognitive bias in volcanic eruption forecasting

Abstract Modern operational eruption forecasting methods rely heavily on human judgment in the face of uncertainty and are thus susceptible to myriad cognitive biases and errors by the scientist-forecasters. Recent developments in the behavioral sciences have elucidated cognitive biases across a wide spectrum of human behaviors and found ways to mitigate them. These insights have led to significant gains in human forecast accuracy across a range of disciplines. However, such gains have yet to widely penetrate volcanic eruption forecasting efforts. In this study, we review recent progress in these fields as relevant to improving current eruption forecasting methods as practiced by volcano observatories worldwide. We group cognitive biases into 1) information and selection biases, 2) group effects and social-emotional biases, and 3) framing and decision biases and highlight the numerous places in which these biases may permeate eruption forecasts. We present a framework for improving group discussion and forecasting processes at volcano observatories, considering recent gains from behavioral sciences along with a mitigation checklist to effectively reduce bias in operational eruption forecasts. Finally, we present an updated forecasting methodology for use in our own group, the US Geological Survey (USGS) Volcano Disaster Assistance Program, based on previous “multiple datasets” methods, that includes clear and deliberate efforts to minimize cognitive biases and thus improve eruption forecasting accuracy.